We study how AI actually works, not just whether it does.
Mechanistic interpretability, principled architectures, and formal reasoning. The questions that shape what intelligence can become.
Flabubium is an early-stage research project focused on the foundations of AI. We work on understanding how models reason, not on building products. The questions we care about do not have obvious timelines, but we think they matter.
Most of the field asks whether something works. We try to ask why. That means studying model internals, circuits, and representations rather than chasing benchmark numbers.
Mechanistic interpretability to understand how models reason. Neural and symbolic architectures built from first principles. Formal theorem proving to bring mathematical rigour to AI.
We are at the start. The goal is to contribute to research that makes AI systems possible to understand and verify. We are a small team working on hard problems, and we are just getting going.